Automated classification of building structures for urban built environment identification using machine learning

Abstract The urban environment, especially buildings with different structure types, is difficult to identify automatically due to variation in urban function and form and to buildings’ relatively tacit structural attributes. To enable automated classification of building structures in complex urban environment, this study developed a machine learning (ML)-based method. Twenty-nine ML features were defined, and twelve popular ML algorithms were tested. The method was tested by classifying over 3700 buildings in Beijing, China into five common structures. Of the 12 ML algorithms, the Gradient Boosted Decision Tree delivered the highest overall performance, and its classification was verified as effective, i.e., achieving approximately 91.7%, 90.6%, and 91.1% of average recall, precision and F1, respectively, on testing data. This study’s classification method helps establish a nexus among urban form, building structures, and materials and resource requirements, thus contributing to the advance of sustainable urban studies, such as urban metabolism and ecological planning.